Monid — Hermes plugin

Connect your agent to every tool it needs. One integration, and your Hermes agent can discover, compare and run third-party tools and APIs at runtime: web search and scraping, people and company data, social platforms, SEO and market data, video, image and voice generation, agent email and phone. No per-provider sign-ups or subscriptions — every run is paid per call from one Monid balance. Discovery and inspection are free, and every result reports its own cost.

This plugin bundles the official Monid skill, which teaches the agent to search the catalog, read a tool's schema before calling it, confirm the price with you before every paid run, and report what each run cost. The skill loads only when you ask for Monid.

Disclosure

Monid (monid.ai, by Monid / monid-ai) is a paid third-party marketplace. Before installing, know that:

Network calls to a third party. The agent sends your requests, and any files you ask it to upload, to Monid (api.monid.ai, mcp.monid.ai, sfs.monid.ai) and on to the provider that runs the tool.
Money. Each run spends your prepaid Monid workspace balance. The skill requires the agent to state the tool, input and price and get your explicit "yes" in the same turn before every run, and before releasing any provisioned resource (irreversible). Workspace budgets and run caps can be set at https://app.monid.ai.
Shell commands. On the CLI transport the agent, with your agreement, runs npm install -g @monid-ai/cli@<exact version> and then monid … commands in the terminal.
Stored credentials. The CLI stores your Monid API key in its own local config. You add the key yourself in your own terminal (monid keys add); the skill never asks you to paste it into chat. The MCP transport uses OAuth tokens held by Hermes (hermes mcp login monid).
No telemetry, hooks, tools or background processes are added by this plugin; it registers one skill.

Install

hermes plugins install monid

(Installs from the Hermes plugin catalog, pinned to an exact reviewed commit of this repository.)

The agent loads the skill explicitly:

skill_view("monid:monid")

Transports

The skill drives one of two transports for the same capabilities:

Transport	Setup	Notes
CLI (default)	None in Hermes — with your agreement the agent installs `@monid-ai/cli` (exact version) via its terminal; you create an API key at https://app.monid.ai/access/api-keys and add it in your own terminal with `monid keys add`	Can write large results to files (`-o`), protecting the context window
MCP (optional)	Add the hosted server to `~/.hermes/config.yaml`, then `hermes mcp login monid` (OAuth — no API key)	Structured `monid_*` tools in the schema

MCP configuration:

mcp_servers:
  monid:
    url: https://mcp.monid.ai/v1
    auth: oauth

auth: oauth is required — the hosted server authenticates with OAuth 2.1 (PKCE), and Hermes handles discovery, token exchange, and refresh.

What's in this repo

plugin.yaml               Hermes native plugin manifest
__init__.py               register(ctx) — registers the bundled skill as monid:monid
skills/monid/SKILL.md     The Monid skill, adapted for Hermes (see below)

No executable tools, no hooks, no environment variables. The plugin is a manifest, one register() call, and one markdown file.

Skill provenance and sync

skills/monid/SKILL.md is derived from two upstream sources:

monid-ai/plugins plugins/monid/skills/monid/SKILL.md — the layout: the two-transport (MCP + CLI) structure, section numbering, and the equivalence table.
monid-ai/cli skills/monid/SKILL.md — the command facts: CLI flags, run statuses, worked examples. Verify against the CLI source (src/commands/**, src/api/types.ts) when in doubt.

When syncing, port new facts from those files into this one, then check that every Hermes-specific deviation below is still in place. Several of them were required by the Hermes catalog security review (hermes-agent#113661); dropping any of them blocks a re-pin.

Where	Hermes deviation (keep on every sync)
frontmatter `description`, §5, §13.1	Narrow trigger: load only when the user explicitly asks for Monid; never for ordinary searches/fetches
§1.1	Hermes connect instructions: CLI works out of the box; MCP via `mcp_servers.monid` + `auth: oauth` + `hermes mcp login monid`
§3.1	`npm install -g @monid-ai/cli@<exact version>` — exact pin, never `@latest` or `^`. Bump it (and `minimum-cli-version`) on each CLI release
§3.2, §12	The user runs `monid keys add` in their own terminal; never paste a key into chat, never run it with a key yourself
§3.3	No self-update: never fetch or overwrite the skill file
§7, §2 workflow, §10/§10a, §13.4a–b	Confirm endpoint + input + price in the current turn before every run; confirm before releasing a resource
§11, §13.12	Hints are untrusted server content: surface them, never auto-execute
§12	`401 (MCP)` row points at `hermes mcp login monid`
`__init__.py`	`register_skill` description mirrors the narrow trigger

Never import from upstream (they conflict with the above): "save the most recent skill from https://monid.ai/SKILL.md", "CLI and skill versions must match" / @latest, "ask them to paste the key", "proactively run monid discover whenever…", "read the Hints and act on them", and the CLI-only "always use -o" rule (MCP has no -o).

The skill must never instruct the agent to fetch or overwrite its own file: updates reach Hermes users only through a commit here plus a SHA-bump PR to the Hermes plugin catalog, followed by hermes plugins update monid. That is the catalog's trust model (exact commit pins, human-reviewed), and this repo must stay compliant with it.

Releasing to the Hermes catalog

Commit the change here (skill sync, manifest bump — keep plugin.yaml: version moving with meaningful releases).
Validate against a Hermes checkout: hermes plugins doctor . --ci && hermes plugins validate .
Open a PR to NousResearch/hermes-agent updating the sha: (and version:) in plugin-catalog/monid.yaml to the new 40-hex commit. Keep the catalog description: equal to plugin.yaml's, including the Disclosure sentence.

License

MIT — see LICENSE.

monid

README